ESTRO 2025 - Abstract Book

S2466

Physics - Autosegmentation

ESTRO 2025

Conclusion: Uncertainty quantification effectively distinguishes ID from OOD data in an auto-segmentation model without prior knowledge of ID/OOD definition. Higher uncertainty, corresponding to lower segmentation performance, was observed in under-represented data, indicating that uncertainty reflects model generalisability. OOD-detection could enhance QA in radiotherapy by flagging cases for manual review.

Keywords: Quality assurance, auto-segmentation, uncertainty

References: [1] Isensee, Fabian, et al. "nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation." Nature methods 18.2 (2021): 203-211. [2] Brouwer, Charlotte L., et al. "CT-based delineation of organs at risk in the head and neck region: DAHANCA, EORTC, GORTEC, HKNPCSG, NCIC CTG, NCRI, NRG Oncology and TROG consensus guidelines." Radiotherapy and Oncology 117.1 (2015): 83-90. [3] Gal, Yarin, and Zoubin Ghahramani. "Dropout as a bayesian approximation: Representing model uncertainty in deep learning." international conference on machine learning. PMLR, 2016